Self-assembled 3D Interconnected Magnetic Nanowire Networks for Neuromorphic Computing.

Abstract

Three-dimensional (3D) nanomagnetic systems offer promise toward implementing neuromorphic computing due to their intricate spin textures, magnetization dynamics, and nontrivial topology. However, the investigation of 3D nanomagnetic systems is often constrained by demanding fabrication and characterization requirements. Here, we present interconnected networks of self-assembled magnetic nanowires (NW) as a novel 3D platform with attractive characteristics for neuromorphic computing. The networks contain multiple unique transport pathways, each hosting discrete magnetization states. These pathways can be selectively addressed, and the magnetic state within them can be electrically controlled by applying current pulses. Consequently, the pathways can serve as synaptic weights, allowing for diverse programming by switching specific sections of the network using current pulses of varying magnitudes and durations. Additionally, unique features such as history-dependent magnetic state switching and interconnected transport paths are observed in these networks. These capabilities are leveraged to illustrate the potential of interconnected magnetic NW networks as reservoir layers in a neural network architecture, highlighting their promise as an efficient platform for neuromorphic computing.